[feat][prompt] prompt as a service (ptaas) (#21)

ptaas

Co-Authored-By: Coda <coda@bytedance.com>

Author: kasarolzzw

.gitignore

@@ -28,4 +28,5 @@ go.work.sum
 /.vscode
 /output
 coverage.out
-/apb
+/apb
+.coda/

client.go

@@ -318,7 +318,8 @@ func GetPrompt(ctx context.Context, param GetPromptParam, options ...GetPromptOp
 
 // PromptFormat format prompt with variables
 func PromptFormat(ctx context.Context, prompt *entity.Prompt, variables map[string]any, options ...PromptFormatOption) (
-	messages []*entity.Message, err error) {
+	messages []*entity.Message, err error,
+) {
 	return getDefaultClient().PromptFormat(ctx, prompt, variables, options...)
 }
 
@@ -495,6 +496,20 @@ func (c *loopClient) PromptFormat(ctx context.Context, loopPrompt *entity.Prompt
 	return c.promptProvider.PromptFormat(ctx, loopPrompt, variables, config)
 }
 
+func (c *loopClient) Execute(ctx context.Context, req *entity.ExecuteParam, options ...ExecuteOption) (entity.ExecuteResult, error) {
+	if c.closed {
+		return entity.ExecuteResult{}, consts.ErrClientClosed
+	}
+	return c.promptProvider.Execute(ctx, req, options...)
+}
+
+func (c *loopClient) ExecuteStreaming(ctx context.Context, req *entity.ExecuteParam, options ...ExecuteStreamingOption) (entity.StreamReader[entity.ExecuteResult], error) {
+	if c.closed {
+		return nil, consts.ErrClientClosed
+	}
+	return c.promptProvider.ExecuteStreaming(ctx, req, options...)
+}
+
 func (c *loopClient) StartSpan(ctx context.Context, name, spanType string, opts ...StartSpanOption) (context.Context, Span) {
 	if c.closed {
 		return ctx, DefaultNoopSpan

entity/prompt.go

@@ -3,7 +3,9 @@
 
 package entity
 
-import "github.com/coze-dev/cozeloop-go/internal/util"
+import (
+	"github.com/coze-dev/cozeloop-go/internal/util"
+)
 
 type Prompt struct {
 	WorkspaceID    string          `json:"workspace_id"`
@@ -29,9 +31,24 @@ const (
 )
 
 type Message struct {
-	Role    Role           `json:"role"`
-	Content *string        `json:"content,omitempty"`
-	Parts   []*ContentPart `json:"parts,omitempty"`
+	Role             Role           `json:"role"`
+	ReasoningContent *string        `json:"reasoning_content,omitempty"`
+	Content          *string        `json:"content,omitempty"`
+	Parts            []*ContentPart `json:"parts,omitempty"`
+	ToolCallID       *string        `json:"tool_call_id,omitempty"`
+	ToolCalls        []*ToolCall    `json:"tool_calls,omitempty"`
+}
+
+type ToolCall struct {
+	Index        int32         `json:"index"`
+	ID           string        `json:"id"`
+	Type         ToolType      `json:"type"`
+	FunctionCall *FunctionCall `json:"function_call,omitempty"`
+}
+
+type FunctionCall struct {
+	Name      string  `json:"name"`
+	Arguments *string `json:"arguments,omitempty"`
 }
 
 type Role string
@@ -45,16 +62,18 @@ const (
 )
 
 type ContentPart struct {
-	Type     ContentType `json:"type"`
-	Text     *string     `json:"text,omitempty"`
-	ImageURL *string     `json:"image_url,omitempty"`
+	Type       ContentType `json:"type"`
+	Text       *string     `json:"text,omitempty"`
+	ImageURL   *string     `json:"image_url,omitempty"`
+	Base64Data *string     `json:"base64_data,omitempty"`
 }
 
 type ContentType string
 
 const (
 	ContentTypeText              ContentType = "text"
 	ContentTypeImageURL          ContentType = "image_url"
+	ContentTypeBase64Data        ContentType = "base64_data"
 	ContentTypeMultiPartVariable ContentType = "multi_part_variable"
 )
 
@@ -119,6 +138,25 @@ type LLMConfig struct {
 	JSONMode         *bool    `json:"json_mode,omitempty"`
 }
 
+type ExecuteParam struct {
+	PromptKey    string         `json:"prompt_key"`
+	Version      string         `json:"version,omitempty"`
+	Label        string         `json:"label,omitempty"`
+	VariableVals map[string]any `json:"variable_vals,omitempty"`
+	Messages     []*Message     `json:"messages,omitempty"`
+}
+
+type ExecuteResult struct {
+	Message      *Message    `json:"message,omitempty"`
+	FinishReason *string     `json:"finish_reason,omitempty"`
+	Usage        *TokenUsage `json:"usage,omitempty"`
+}
+
+type TokenUsage struct {
+	InputTokens  int `json:"input_tokens"`
+	OutputTokens int `json:"output_tokens"`
+}
+
 func (p *Prompt) DeepCopy() *Prompt {
 	if p == nil {
 		return nil
@@ -181,12 +219,14 @@ func (cp *ContentPart) DeepCopy() *ContentPart {
 		return nil
 	}
 	copied := &ContentPart{
-		Type:     cp.Type,
-		ImageURL: cp.ImageURL,
+		Type: cp.Type,
 	}
 	if cp.Text != nil {
 		copied.Text = util.Ptr(*cp.Text)
 	}
+	if cp.ImageURL != nil {
+		copied.ImageURL = util.Ptr(*cp.ImageURL)
+	}
 	return copied
 }
 

entity/stream.go

@@ -0,0 +1,8 @@
+// Copyright (c) 2025 Bytedance Ltd. and/or its affiliates
+// SPDX-License-Identifier: MIT
+
+package entity
+
+type StreamReader[T any] interface {
+	Recv() (T, error)
+}

error.go

@@ -16,5 +16,7 @@ var (
 	ErrParsePrivateKey  = consts.ErrParsePrivateKey
 )
 
-type AuthError = consts.AuthError
-type RemoteServiceError = consts.RemoteServiceError
+type (
+	AuthError          = consts.AuthError
+	RemoteServiceError = consts.RemoteServiceError
+)

examples/init/log/log.go

@@ -31,8 +31,7 @@ func main() {
 	cozeloop.Close(ctx)
 }
 
-type CustomLogger struct {
-}
+type CustomLogger struct{}
 
 func (l *CustomLogger) CtxDebugf(ctx context.Context, format string, v ...interface{}) {
 	fmt.Printf("[Custom] [DEBUG] "+format+"\n", v...)

examples/prompt/prompt_hub.go examples/prompt/prompt_hub/prompt_hub.goexamples/prompt/prompt_hub.go renamed to examples/prompt/prompt_hub/prompt_hub.go

@@ -131,8 +131,8 @@ func (r *llmRunner) llmCall(ctx context.Context, messages []*entity.Message) (er
 	defer span.Finish(ctx)
 
 	// llm is processing
-	//baseURL := "https://xxx"
-	//ak := "****"
+	// baseURL := "https://xxx"
+	// ak := "****"
 	modelName := "gpt-4o-2024-05-13"
 	maxTokens := 1000 // range: [0, 4096]
 	//transport := &MyTransport{

…mpt/prompt_hub_jinja/prompt_hub_jinja.go …hub/prompt_hub_jinja/prompt_hub_jinja.goexamples/prompt/prompt_hub_jinja/prompt_hub_jinja.go renamed to examples/prompt/prompt_hub/prompt_hub_jinja/prompt_hub_jinja.go examples/prompt/prompt_hub_jinja/prompt_hub_jinja.go renamed to examples/prompt/prompt_hub/prompt_hub_jinja/prompt_hub_jinja.go

@@ -160,8 +160,8 @@ func (r *llmRunner) llmCall(ctx context.Context, messages []*entity.Message) (er
 	defer span.Finish(ctx)
 
 	// llm is processing
-	//baseURL := "https://xxx"
-	//ak := "****"
+	// baseURL := "https://xxx"
+	// ak := "****"
 	modelName := "gpt-4o-2024-05-13"
 	maxTokens := 1000 // range: [0, 4096]
 	//transport := &MyTransport{

…mpt/prompt_hub_label/prompt_hub_label.go …hub/prompt_hub_label/prompt_hub_label.goexamples/prompt/prompt_hub_label/prompt_hub_label.go renamed to examples/prompt/prompt_hub/prompt_hub_label/prompt_hub_label.go examples/prompt/prompt_hub_label/prompt_hub_label.go renamed to examples/prompt/prompt_hub/prompt_hub_label/prompt_hub_label.go

@@ -125,8 +125,8 @@ func (r *llmRunner) llmCall(ctx context.Context, messages []*entity.Message) (er
 	defer span.Finish(ctx)
 
 	// llm is processing
-	//baseURL := "https://xxx"
-	//ak := "****"
+	// baseURL := "https://xxx"
+	// ak := "****"
 	modelName := "gpt-4o-2024-05-13"
 	maxTokens := 1000 // range: [0, 4096]
 	//transport := &MyTransport{

…pt_hub_multipart/prompt_hub_multipart.go …pt_hub_multipart/prompt_hub_multipart.goexamples/prompt/prompt_hub_multipart/prompt_hub_multipart.go renamed to examples/prompt/prompt_hub/prompt_hub_multipart/prompt_hub_multipart.go examples/prompt/prompt_hub_multipart/prompt_hub_multipart.go renamed to examples/prompt/prompt_hub/prompt_hub_multipart/prompt_hub_multipart.go

@@ -77,7 +77,7 @@ func main() {
 
 		// 4.Format messages of the prompt
 		imageText := "图片样例"
-		imageURL := "https://example.com" //公网访问地址
+		imageURL := "https://example.com" // 公网访问地址
 		messages, err := llmRunner.client.PromptFormat(ctx, prompt, map[string]any{
 			"num":   "2",
 			"count": 10,
@@ -131,8 +131,8 @@ func (r *llmRunner) llmCall(ctx context.Context, messages []*entity.Message) (er
 	defer span.Finish(ctx)
 
 	// llm is processing
-	//baseURL := "https://xxx"
-	//ak := "****"
+	// baseURL := "https://xxx"
+	// ak := "****"
 	modelName := "gpt-4o-2024-05-13"
 	maxTokens := 1000 // range: [0, 4096]
 	//transport := &MyTransport{